llama.cpp
ggml : optimize cuda cumsum fallback (~2.5x speedup vs CUB)
#18343

Merged

ggml : optimize cuda cumsum fallback (~2.5x speedup vs CUB) #18343

am17an merged 1 commit into ggml-org:master from Aadeshveer:ggml-cuda-cumsum-opt

github-actions added Nvidia GPU

github-actions added ggml

am17an commented on 2025-12-24

Aadeshveer force pushed 269 days ago

Aadeshveer requested a review from

am17an 269 days ago

Aadeshveer force pushed 269 days ago

am17an approved these changes on 2025-12-24

ggml : optimize cuda cumsum fallback kernel

4bc188b9

Aadeshveer force pushed to 4bc188b9 269 days ago

am17an merged c54bba86 into master 268 days ago

Reviewers

am17an

Assignees

No one assigned

Labels

Nvidia GPU ggml

Milestone

No milestone